The Impact of Cardiovascular Disease and Chronic Kidney Disease on Life Expectancy and Direct Medical Cost in a 10-Year Diabetes Cohort Study

Diabetes is a highly prevalent noncommunicable disease, affecting 451 million people and causing 5 million deaths worldwide in 2017 (1). It is estimated that the number of patients with diabetes will rise to 693 million by 2045 (1). The annual global medical costs for patients with diabetes was estimated to be ∼$612-$1,099 billion (U.S. dollars) in 2014 (2). The average medical cost spent on each patient with diabetes was 2.3 times higher than on those patients without diabetes (3). Cardiovascular diseases (CVDs) and chronic kidney diseases (CKDs) are two major causes of morbidity, affecting 20–40% of patients with diabetes (4–6). The number of CVD incidences increased by >25% during the period from 1990 to 2010, whereas that of CKD doubled in the general population (7). Given that life expectancy is increasing, the prevalence of CVD and CKD among all patients with diabetes will keep growing, and thus increase the economic burden as well as premature mortality.

Some studies have suggested that the mortality risks for CVD and CKD are similar in the general population (8,9). Nevertheless, most studies conducted among patients with diabetes have assessed the mortality risk caused by either CVD or CKD alone (10–14). Moreover, the results to date have been inconsistent in that the relative mortality risks have ranged from 1.5 to 3.3 for CVD and from 0.94 to 5.0 for CKD (10–15). It is therefore hard to compare the differences between the burden of the CVD and CKD. The impact of the comorbid CVD and CKD is also unclear, such as whether the effect on CVD and CKD is additive or multiplicative remains unknown. There is currently only limited evidence demonstrating the actual reduction of life expectancy among these patients, which is more clinically meaningful and easy to understand. In addition, the medical cost for treating CVD and CKD among patients with diabetes has been inconsistent between studies (16,17). The most appropriate method to evaluate these associations is to estimate the burden of CVD and CKD in the same cohort. Hence, the aim of this study was to evaluate the impact of CVD and CKD individually and jointly on mortality risks, life expectancy, and direct medical costs based on a 10-year Hong Kong diabetes cohort. This will help facilitate the development of health care policies, including resource allocation and treatment prioritization for the prevention of CVD and CKD for patients with diabetes.

This population-based retrospective cohort study included patients aged 18 years or older with the diagnosis of diabetes and managed by the Hong Kong Hospital Authority over the period between 1 January 2008 and 31 December 2010. The Hong Kong Hospital Authority is the statutory body that manages all of the public-sector hospitals and primary care clinics in Hong Kong. It is responsible for managing >90% of patients with diabetes because the health services are heavily subsidized by the Hong Kong government (18). The ecology of Hong Kong health care system is comparable to the U.S. and U.K. The monthly rates for outpatient care used in Hong Kong are higher compared with the U.S. and U.K., but the use of hospital-based events, including accident and emergency attendances and hospitalizations, is similar (19). The International Classification of Primary Care-2 (ICPC-2) codes of “T89” or “T90” are used to define the clinical diagnoses of DM. Baseline was defined as the first attendance of a doctor consultation in a primary care outpatient clinic during the inclusion period.

Ethics approval was received from the Institutional Review Boards of the Hong Kong Hospital Authority (Ref: UW 15-259). Consent from individual subjects was not needed because all patient records were retrieved from the computerized administrative system of the Hospital Authority anonymously.

Subjects were divided into 12 mutually exclusive disease and disease combination groups at the baseline: 1) none of these (reference group); 2) stroke; 3) heart diseases, including coronary heart disease and heart failure; 4) moderate CKD; 5) severe CKD; 6) stroke and heart diseases; 7) heart diseases and moderate CKD; 8) stroke and moderate CKD; 9) stroke and severe CKD; 10) heart diseases and severe CKD; 11) stroke, heart diseases, and moderate CKD; 12) stroke, heart diseases, and severe CKD. The diagnoses of CVD and CKD were defined according to the diagnostic codes of the ICPC-2 or the International Classification of Diseases, Ninth Clinical Modification. Moderate CKD was defined as estimated glomerular filtration rate (eGFR) of ≥30 mL/min/1.73 m² and <60 mL/min/1.73 m² and severe CKD as eGFR of <30 mL/min/1.73 m² (as described in Supplementary Table 1).

The primary outcome was all-cause mortality. Mortality data were obtained from the Hong Kong Death Registry, a population-based official government registry with the registered death records of all Hong Kong citizens. Meanwhile, secondary outcomes were the annual direct medical cost and life expectancy. The total medical cost per patient was calculated by the product of the frequency of attendance and the unit cost of attendance. It included the use of medical services, such as general and specialist outpatient clinics, allied health professional services (including clinical psychologists, dietitians, occupational therapists, and physiotherapists), accident and emergency services, and hospital inpatient services. The frequency for health service use was retrieved from the Hong Kong Hospital Authority computerized administrative system. The unit cost of relevant health service use was based on the published costs according to the Government of the Hong Kong Special Administrative Region Gazette and Hospital Authority Ordinance (chapter 113) in 2013 (20). The unit cost is based on a cost recovery basis and is the lump sum covering all medical services used during the visits, including consultation, investigations, medications, and other treatments used. The annual public direct medical costs referred to the total cost for each kind of health service attendance.

The patients’ sociodemographics, behavior characteristics, clinical parameters, disease characteristics, and treatment modalities are the baseline covariates. Sociodemographics parameters include sex and age. Behavior characteristics include smoking status. Clinical parameters consist of systolic and diastolic blood pressure (SBP and DBP), LDL-cholesterol (LDL-C), hemoglobin A_1c (HbA_1c), and BMI. Disease characteristics include self-reported duration of diabetes medication treatment, consisting of the use of antihypertensive drugs, including ACE inhibitor/angiotensin receptor blockers, β-blockers, calcium channel blockers, diuretics, and other antihypertensive drugs, the use of antidiabetes drugs, and the use of lipid-lowering agents at baseline. All laboratory tests were conducted in Hospital Authority laboratories using the same protocol and accredited by the College of American Pathologists, the Hong Kong Accreditation Service, or the National Association of Testing Authorities, Sydney, New South Wales, Australia.

Multiple imputation was used to handle all missing baseline covariates to minimize potential bias influenced by missing data (21). The chained equation method was used to impute all the missing data five times with all baseline covariates and mortality outcome. Pooled estimates and the corresponding 95% CIs were calculated based on the Rubin rule (22). Descriptive statistics were used to summarize patients’ characteristics for each disease status group. Cumulative incidences and incidence rates for all-cause mortality with 95% CIs were calculated. Incidence rates for outcome events were estimated based on the 95% CI under Poisson distribution (23). To evaluate the associations between disease group status and primary outcome, multivariable Cox proportional hazards regression models were used with adjustments based on all baseline characteristics. Proportional hazards assumptions were also considered while fitting the multivariable Cox proportional hazards regression models by using the scaled Schoenfeld residuals plots against time for the covariates. The variance inflation factor was used to determine the presence of multicollinearity. Hazard ratios (HRs) with corresponding 95% CIs and P values were reported accordingly. Three measurements, 1) the relative excess risk due to interaction (RERI), 2) the attributable proportion due to interaction (AP), and 3) the synergy index (SI), were used to examine the additive or multiplicative interaction between disease status groups based on the adjusted HRs from the Cox regression. RERI and AP equaling 0 and SI equaling 1 indicated the absence of an interaction effect (24) and suggested an additive interaction between disease status groups.

Subgroup analyses were used to determine whether there were any relationships between different disease status groups and mortality. Stratification was by sex (men, women), age (<65 years, ≥65 years), smoking status (nonsmoker, smoker), duration of diabetes (<5 years, ≥5 years), BMI (<27.5 kg/m², ≥27.5 kg/m²), blood pressure (SBP <130 mmHg and DBP <80 mmHg, SBP/DBP ≥130/80 mmHg), HbA_1c (<7%, ≥7%), LDL-C (<2.6 mmol/L, ≥2.6 mmol/L), the use of antihypertension drugs (no, yes), the use of antidiabetes drugs (no, yes), and the use of lipid-lowering drugs (no, yes) at baseline.

For secondary outcomes, the estimated direct medical cost incurred for each disease status group per year was calculated using the mean of the annual cost of health service attendances within 5-years after baseline. A generalized linear method with gamma family, log-link function, and adjusting for baseline characteristics was used to evaluate the adjusted difference in annual cost of public health service use among disease status groups. This model could be used in the application of estimating the additive effect of complications on medical costs (17,25,26). Meanwhile, a flexible parametric survival model for relative survival was used to estimate the life expectancy losses within different disease status groups (27). This model could calculate the life expectancy (28–30) based on the measurement of loss in expectation of life by extrapolating the estimated linear trend at the end of the follow-up period without consideration of any proportional hazards model assumption. Age and sex were considered as covariates with time-dependent effects. Restricted cubic splines were used to model age continuously and nonlinearly.

All significance tests were two-tailed and considered statistically significant if P values were <0.05. Statistical analysis was performed with Stata version 13.0 software.

Owing to the confidentiality of the data used for this study and strict privacy policy from the data holder that the data can be kept among the designated research personnel only, the data cannot be provided to others, whether or not the data are made anonymous.

At total of 208,792 patients with diabetes were identified and included for analysis in this cohort study. Of these, 11,922 (5.7%), 10,736 (5.1%), 11,367 (5.4%), and 1,114 (0.5%) had stroke, heart diseases, moderate CKD and severe CKD, respectively. Among all patients with diabetes, 35,139 patients (16.8%) had one condition of stroke, heart disease, or CKD, 6,981 (3.3%) had two conditions, and 801 (0.4%) had three conditions. Supplementary Table 2 summarizes the data completion rates for each baseline covariate (with >75% data completion on average). Table 1 reports the baseline characteristics of the cohort by disease group status. In general, the average age of the subjects was 65 (SD 12) years old, where women accounted for 54% of the cohort.

After a median follow-up period of 8.5 years (1.6 million patient-years), there were 50,154 deaths. Table 2 reports the incidence rate for mortality ranged from 2.11 to 40.95 per 100 patient-years among the different disease status groups. The trend in the incidence rate increased from one, two, and three conditions among stroke, heart, and moderate CKD. In particular, the incidence rate for mortality in patients with severe CKD was higher than any combination of stroke, heart, and moderate CKD. A similar pattern of HRs was observed after adjusting for other covariates, as reported in Table 2. The mortality risk was higher for patients with more conditions among CKD, heart disease, and stroke. Compared with those without a history of CKD and CVD, the adjusted HRs for subjects with one, two, and three conditions among stroke, heart disease, and moderate CKD was increased by 1.75 times, 2.63 times, and 3.58 times, respectively. The mortality risk for patients with severe CKD was much higher. Supplementary Table 3 demonstrates that RERI, AP, and SI for stroke, heart disease, moderate/severe CKD, except for stroke, and severe CKD absented the interaction effect, which suggests that it is more likely that there is an additive interaction between disease status groups. Supplementary Table 4 shows comparable patterns in different subgroups compared with the main analysis.

Table 3 summarizes the estimated public direct medical costs per annum, ranging from $4,065 to $35,362 (U.S. dollars) across the different disease status groups. A similar trend was observed in the mortality rates, with an incremental increase in medical costs with a cumulative increase in number of conditions. The annual public medical cost for patients with diabetes with one, two, and three conditions among stroke, heart disease, and moderate CKD increased by 1.91 times, 2.90 times, and 3.88 times compared with those without a history of CKD and CVD respectively. The medical costs for patients with severe CKD alone or in a combination of stroke and heart disease was much higher. More detailed findings are provided in Table 3.

Figure 1 shows the estimated loss of life expectancy for different disease status groups (with comparison group). The estimated life expectancy loss for men (women) aged 40 years old with stroke, heart diseases, and moderate CKD was nearly 21.1 years (95% CI 15.5–26.6) (women 26.2 years [95% CI 20.6–31.9]), 16.6 years (95% CI 9.6–23.7) (women 21.7 years [95% CI 14.0–29.5]), 23.2 years (95% CI 18.6–27.7) (women 28.3 years [95% CI 23.7–32.9]), and 31.5 years (95% CI 30.2–32.7) (women 36.3 years [95% CI 35.0–37.7]), respectively. In addition, the estimated life expectancy loss for men (women) with severe CKD or in a combination of any two or three conditions was even more prominent, with the results shown in Fig. 1. The life expectancy loss for patients of younger age was much larger than that of older patients, but the reduction in life expectancy for elderly was still high. For instance, the years of life lost for 40-year-old men (women) with diabetes with stroke, heart disease, and severe CKD was 18.9 (95% CI 18.2–19.6) (women 22.6 [95% CI 21.9–23.4]).

This large cohort study involved >50,000 incident deaths across 1.59 million patient-years at risk and contributes to the evidence for estimating mortality risk, direct medical costs, and life expectancy loss due to various combinations of CVD and CKD for patients with diabetes. The findings demonstrated similar mortality risks associated with stroke, heart diseases, and moderate CKD. This supports the recommendation from the National Kidney Foundation and the American College of Cardiology/American Heart Association that CKD should be treated as a CHD risk equivalent (31,32). Our analyses also found the effect of stroke, heart diseases, and moderate CKD on mortality risk were individually additive and nonoverlapping for any combination. The mortality risk associated with severe CKD is larger than the combined effects of stroke, heart diseases, and moderate CKD. Similar patterns were observed for direct medical costs. Additionally, increases in the number of conditions reduced life expectancy greatly, particularly among younger patients.

No previous study has yet evaluated the individual and combined effect of CVD and CKD on mortality risk and direct medical cost among DM patients. A few studies conducted in the general population have found that patients with stroke, heart diseases, and CKD had similar mortality risks and direct medical costs (33,34). Our study confirmed this finding and extended the impact of these combinations of diseases on mortality risks and direct medical cost in patients with DM. In this cohort of patients with DM, the risks were observed to be additive and nonoverlapping. The increase in the number of conditions, including stroke, heart disease, and CKD, corresponded to the rise in mortality risks and direct medical costs. Our findings provide further evidence for the importance of CVD prevention among patients with CKD and the prevention of CKD among patients with CVD (31,35,36) and suggest that prevention of CVD and CKD might play equally important roles in decreasing the disease burden in patients with diabetes. A previous study conducted in the Canadian general population demonstrated similar mortality risks for patients with CHD and those with moderate to severe CKD (9). Our results partially confirmed this finding because of the substantially higher mortality risk associated with severe CKD compared with coronary heart disease (CHD) among patients with diabetes. We recommend that the effect of severe CKD and moderate CKD should be considered independently when evaluating mortality risks and that prevention of severe CKD be prioritized.

The data from the 1950–1980s Framingham Heart Study with 9,033 participants demonstrated that men and women aged 50 years with diabetes and CVD had 7.1 years’ and 6.8 years’ reduction in life expectancy, respectively, compared with those without CVD (37). The results were lower than our expectation. One potential explanation could be the rapid increase in global life expectancy between 1950 and 2010 (38). Although ethnic differences and the historical characteristics of the Framingham Heart Study discourage extrapolation of those findings to the current context, our sample size was much larger, and more detailed analytical techniques were used, which can provide more reliable results. Meanwhile, two large, general population cohort studies demonstrated a reduction of life expectancy for Taiwanese people aged 60 with moderate and severe CKD was ∼4–13 years but was 4–9 years among Canadian subjects (33). These years of life loss results were less than those observed in our current study because our study population was limited to patients with diabetes. Patients with diabetes have an average of ∼8–10 years life expectancy loss compared with those without diabetes (14,37). Hence, our findings would likely be similar to the previous studies if diabetes was taken into account. The average life expectancy loss for men and women with stroke, heart diseases, and severe CKD within the age-group of 40–60 years (decrease by 37 years and 21 years, respectively) were surprising in that they were larger than the 10-year loss for lifelong smokers and 11-year loss for those with HIV infection (39–41).

Our findings demonstrated that ∼20% of our cohort of patients with diabetes had at least one complication among CKD, stroke, and heart diseases. Limited symptoms could be observed at the earlier stages of CKD because of the difficulty in discovering the symptoms related to uremia until the latest stage with severe CKD (42). Therefore, the early stages of CKD could gradually progress to severe CKD asymptomatically. Previous studies in Taiwan showed an underdiagnosis, undertreatment, and lack of awareness for CKD (43). A review suggested that early treatments, including an education program and pharmacologic therapies, might help to reverse the early stages of CKD (44). In addition, most treatment guidelines currently focus primarily on CVD risk screening, with care plans stratified by risk. Even though various common risk factors are recognized for both CKD and CVD, proper screening for early identification and management for CKD could potentially slow down or even prevent CKD deterioration. While many clinicians focus on the importance in controlling and preventing CVD, the severity and prevention of CKD may be easily neglected. Our findings highlight the importance of CKD management for patients with diabetes directed at preventing and delaying CKD deterioration.

One major strength of this cohort study was the large sample size of patients with diabetes monitored in primary care. All missing data handling was conducted using multiple imputation and multiple adjustment with some confounding variables included to show the comparison of burdens between CKD and CVD. Using data extracted from the Hong Kong Hospital Authority’s administrative database ensured that the patient data were accurate and reliable.

There are also several potential limitations. Only patients with diabetes in Hong Kong were included in our study. Mortality risks, direct medical cost, and health policies vary with time, and thus, our findings might be inappropriate to extend to the general population and to patients with diabetes living in other countries or regions. Researchers should be cautious if extrapolating our study findings to other settings. Some potential confounding variables, such as drug compliance and use and socioeconomic characteristics, were not considered in this study. Duration of diabetes was self-reported and liable to self-report biases. However, our data analysis included the main clinical parameters and different medications. Approximately 1% of the patients included in this cohort had a diagnosis of type 1 diabetes. Owing to the small sample size, our analyses were not able to be stratified by diabetes type. Hence, the results in study are largely only generalizable to patients with type 2 diabetes. The ICPC-2 codes were used to identify patients with diabetes because International Classification of Diseases, Ninth Clinical Modification codes for diabetes were unavailable in our data set, and thus, our cohort missed patients who were not coded with an ICPC-2 diabetes code. Our multiple imputation model did not include the medical costs for all missing predictors. Lastly, further studies with longer follow-up periods are required to make an accurate estimation of the projection of life expectancy.

This cohort study found high mortality risks and large annual public direct medical cost associated with the burden of with severe CKD among patients with diabetes. Our findings demonstrated similar burdens were incurred by those with a history of moderate CKD, stroke, and heart diseases, which were closely additive and nonoverlapping for different disease combinations. The estimated life expectancy of patients with diabetes with CKD and CVD drops gradually as disease severity deteriorates. Our findings highlight the importance of long term CKD and CVD management and prevention in patients with diabetes.

Acknowledgments. The authors wish to acknowledge the contributions of the Hong Kong Hospital Authority.

Funding. The study received support from the Health Services Research Fund, Food and Health Bureau, Hong Kong Special Administrative Region (ref. no. 14151181).

No funding organization had any role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; or preparation of the manuscript.

Duality of Interest. I.C.K.W. received funding from Pfizer, Bayer, and Novartis to evaluate real-world evidence on pharmacological treatments of CVDs but not related to the current study. E.W.Y.C. received research grants from Bayer, Bristol-Myers Squibb, Janssen (a Division of Johnson and Johnson), Pfizer, and Takeda to evaluate real-world evidence on pharmacological treatments of CVDs but not related to the current study. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. E.Y.F.W. and C.L.K.L. contributed to the study design and acquisition of data, researched the data, contributed to the statistical analysis and interpretation of the results, and wrote the manuscript. W.Y.C. wrote the manuscript, contributed to the statistical analysis and interpretation of the results, and reviewed and edited the manuscript. All authors contributed to the interpretation of the results and reviewed and edited the manuscript. E.Y.F.W. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.